Sources: Uber's autonomous cars struggled with goal of 13 miles before driver intervention; Waymo says its cars had average of ~5.6K miles between interventions
versus nearly 5,600 for Waymo and 1,200 for Cruise in CA. Uber says MPI is not a good measure of safety. However, it is a measure of performance reliability Daisuke Wakabayashi / @daiwaka : That's a fair question. But I must say that i felt like they were very honest and upfront when it came to this story. We didn't agree on the interpretation of certain things, but I never got the sense that they were being dishonest. https://twitter.com/... Daisuke Wakabayashi / @daiwaka : Before people get worked up, I'm not saying that this could have prevented the fatal crash. But it's kinda crazy to me that companies are left on their own to decide HOW they test cars even thought they are on public roads. Johana Bhuiyan / @jmbooyah : So we reported almost exactly a year ago Uber drivers had to intervene once per .8 miles. https://www.recode.net/... New docs that shows they're now struggling to get to 1 every 13 miles. This is still very far behind competitors and not a ton of progress https://twitter.com/... Daisuke Wakabayashi / @daiwaka : Here's my overall take: It's kind of nuts that we are letting people beta test this stuff on the roads without specific rules for how they can test this stuff. I agree that eventually driverless will be safer than humans, but let's not assume it is today Daisuke Wakabayashi / @daiwaka : Uber's testing was working up to lauching a driverless car service in Phoenix under Uber X by year-end. Daisuke Wakabayashi / @daiwaka : From the cutting room floor: Uber had focused its Phoenix efforts around mastering two main routes. One was a loop around ASU and another — called the Scotsdale canonical — was the route where Elaine Herzberg was struck on Sunday night.
Context & Ripple Effects
A year after internal documents showed Uber's 43-car fleet needing human intervention every 0.8 miles with 'critical interventions' every 200 miles (Recode's review of Uber's internal testing logs), the NYT reports the company was still struggling toward a goal of just 13 miles between interventions. Against that, Waymo claims roughly 5,600 miles per intervention and Cruise about 1,200 in California — a two-orders-of-magnitude gap at the top of the table.
The timing matters: Uber was testing toward launching a driverless service on Uber X in Phoenix by year-end, and the reporting lands amid scrutiny following the fatal Tempe-area crash that killed Elaine Herzberg. Uber's response — that miles per intervention (MPI) is not a good measure of safety — sets up a fight over which metric regulators and the public should trust.
First-order effects
- Uber's planned driverless Uber X launch in Phoenix is directly exposed: a 13-mile intervention target against Waymo's ~5,600 makes its readiness case hard to defend to regulators, partners, and riders.
- Waymo and Cruise now have their MPI figures circulating as public benchmarks, whether or not they chose the comparison — Cruise's ~1,200 miles shows even the second-best disclosed number dwarfs Uber's.
Second-order effects
- Uber's pushback that MPI measures performance reliability rather than safety forces the question of what metric should govern — pushing rivals and regulators toward standardized, auditable safety reporting instead of self-selected numbers.
- The disclosure gap becomes competitive ammunition: Waymo can point to a four-hundred-fold spread when courting the same city launches, investors, and talent that Uber's program needs to survive post-crash.
Third-order effects
- If MPI-style disclosures become the de facto yardstick, autonomous programs will be ranked publicly on intervention rates, pressuring laggards to either publish comparable data or cede the driverless-launch race — an accountability structure the industry has so far avoided.
- The pattern points toward regulator-mandated safety auditability for AV testing, replacing today's voluntary, contested metrics with uniform reporting requirements across states.
The trend: Autonomous vehicle competition is shifting from who tests where to who can prove reliability with auditable metrics, and Uber's intervention-rate gap shows how quickly disclosure itself becomes the battleground.